pith. sign in

Fast and Efficient Parallel Sampling Using Higher Order Langevin Dynamics

1 Pith paper cite this work. Polarity classification is still indexing.

1 Pith paper citing it
abstract

We study parallel sampling from high-dimensional strongly log-concave distributions. Langevin-based samplers converge rapidly in continuous time, but their discretizations are typically sequential and often require polynomially many steps in the dimension $d$, the target accuracy $\varepsilon^{-1}$, or both. Picard-based parallel sampling methods reduce this sequential depth to polylogarithmic scale by solving for many time-discretization points in parallel; however, existing guarantees often require a polynomial number of processors, leading to substantial memory and gradient-evaluation costs in high dimensions. We show that higher-order Langevin structure can reduce this parallel resource burden while preserving polylogarithmic sequential depth. Our method combines arbitrary-order Langevin dynamics with blockwise Lagrange polynomial interpolation. This sharper discretization reduces the number of parallel points required to achieve a target accuracy. Our results cover both higher-order smooth potentials and ridge-separable potentials, including models such as Bayesian logistic regression and two-layer neural networks, and improve upon the space complexity of the current literature on parallel log-concave sampling.

fields

cs.DS 1

years

2026 1

verdicts

UNVERDICTED 1

clear filters

representative citing papers

Smoothed Score Queries and the Complexity of Sampling

cs.DS · 2026-05-26 · unverdicted · novelty 8.0

Smoothed-score queries to the resolvent of a Gaussian precision matrix yield an O((log κ) log(1/δ)) query sampler for TV error δ and an Ω(log κ) bit lower bound, improving the condition-number dependence from √κ.

citing papers explorer

Showing 1 of 1 citing paper after filters.

  • Smoothed Score Queries and the Complexity of Sampling cs.DS · 2026-05-26 · unverdicted · none · ref 21 · internal anchor

    Smoothed-score queries to the resolvent of a Gaussian precision matrix yield an O((log κ) log(1/δ)) query sampler for TV error δ and an Ω(log κ) bit lower bound, improving the condition-number dependence from √κ.